The use of spatial expressions always involves some degree of ambiguity. Some of the ambiguity is quantitative. For example, when someone says, “the ball is to the LEFT of the yellow car”, it is not obvious how we can determine the area which is “the LEFT of” the car. Where in the space does “LEFT” end and “FRONT” start? Other type of amibiguity is qualitative. In the above example, we even cannot assume that the speaker and the hearer have the same spatial relation in mind. Did the speaker say LEFT meaning LEFT with respect to the speaker , or LEFT with respect to the car's front ? Two different perspective systems underlie the semantics of basic spatial terms such as FRONT/BACK/LEFT/RIGHT. One system uses the viewer as the frame of reference (the deictic system), while the other system takes one object as the frame of reference and describes other objects with respect to this reference object (the intrinsic system). As shown in the “to the left of the car” example above, these two systems yield very different interpretation of the spatial layout in some situations. Although this fact has been long noted by many researchers (e.g., Clark, 1973; Fillmore, 1975; Levelt, 1982; Levinson, 1996), how people select one system over the other under various circumstances still needs to be explored (Franklin, Tversky, & Coon, 1992). In this research, by setting up an experiment using a 3-D model space in a computer display, we empirically examined how people determined the assignment of FRONT/BACK/LEFT/RIGHT in various contexts including different types of reference objects as well as different orientations these objects were placed with respect to the perspective of the viewer. We found that at least four factors, (1) characteristics of the reference object; (2) orientation of the reference object with respect to the viewer; (3) position of the referred object with respect to the reference object; (4) mode of presentation, are involved when people select one perspective system over the other, and furthermore, there is a complex interaction among these factors. We also examined the boundaries of the four terms. We found that the four terms did not evenly divide the given space. BACK in general covered a larger area than the other three terms, but the division of the space by the four terms are dynamically affected by the orientation of the reference object.